Instalment Detector
SUMMARY
- Client
- Raiffeisenbank
- Tech stack
- Jupyter, Hive, Apache Spark, Hadoop, Python
The majority of the profit in modern retail banking comes from loan products. Unfortunately, not all of your clients borrow money from you. Our data science model is built to detect all payments going from the bank to other banks and financial institutions to pay for existing loans elsewhere. The whole instalment loan battlefield can be laid open in front of your eyes. This is a perfect opportunity to offer clients a loan transfer or consolidation through a bank.
Results
Project Background
Our client, Raiffeisenbank CZ, offers an outstanding online service and is continuously pushing out new products to meet customer needs and expectations.
The bank was keen to maintain its position in the industry, so we looked at how we could use state-of-the-art technology to help them.
Working alongside our client, we asked the question, “How can we use this collected data fully, in a way that enables us to offer bank customers an even better service?” Our answer came via applying advanced data analytics and machine-learning methods to the problem – enabling us to make competitive loan consolidation offers to the right customers.
Challenge
Detecting loan instalments paid to other lenders, within customer transactional data, involves executing complex computations over hundreds of millions of records on a daily basis. A robust big data pipeline for high parallel data processing is needed, as well as the inclusion of suitable data science tools and methodology.
It is cutting-edge work. In fact, this project was the very first implementation of this kind into the bank environment, without any existing technological or architecture blueprint.
Business Needs
The solution needed to meet the following specifications:
- Identify clients with competitor loans for targeted marketing campaigns focused on consolidation
- More accurate assessment of customers’ credit risk scoring
- Adding data science tools to the bank infrastructure and setting up a big data processing pipeline
- High-performance technology to promptly process clients’ transactions without delays
Solution & Results
We designed a complex processing pipeline, implemented on a local Hadoop cluster, including data science tools such as Apache Spark, Hive, and Jupyter. In order to identify customers with loans elsewhere, we applied our instalment detection tool.
Using an instalment detection tool
The tool processes customers’ banking transactions and related data. It’s calibrated specifically to automatically detect loan instalments for each customer. The model is based on advanced statistical and machine-learning methods such as Multi-layer Bayesian Networks. Implementation into the big data pipeline means it can handle processing huge volumes of transactional data – even billions of records on a daily basis.
HEAR FROM THE CLIENT
As a part of the solution, Profinit implemented a tool for automatic detection of external loans. This instalment detection tool uncovers customers’ obligations to other banking and non-banking institutions, which brings valuable information for possible offers of loan consolidation, better-targeted marketing and more accurate credit risk scoring.
